小雪带你了解汽车

终于成为有车一族了分享下提车流程｜购车分享 @marilyn_1003

[wast3 @0xWast3]

Anthropic engineering team stopped letting models touch the app and gave them a script layer instead, and the whole thing got faster overnight

A model holding the tool directly has to understand the tool. A model writing a script only has to understand the job.

> READ - the task parsed once into what has to change, never into UI steps
> WRITE - the model emits a script instead of clicking through an interface
> RUN - the script executes natively, at the speed the software actually works
> CHECK - the result verified against the task, not against what the model intended
> KEEP - scripts that worked saved, so the same job never gets re-reasoned

Direct tool use looks impressive and runs slowly. Every action is a round trip: look, decide, act, look again.

A script collapses all of that into one pass. The thinking happens once, the execution happens at machine speed.

The keep step is where it compounds. The second time a job appears, there is nothing to generate at all.

Verification against the task matters more than it sounds. A script that runs cleanly can still do the wrong thing confidently.

Most agent stacks are slow for this exact reason. Not weak models, just models forced to operate software one click at a time.

The script layer, the verification rules and a full before-and-after run are laid out underneath.

[小雪带你了解汽车 @Maya_Jey]

@0xWast3 the removal and cleansing effort deserved a running timer

[meya. @birmeya]

https://t.co/X1mTgC6DRv

[Orion @0xOrionVega]

this is getting f*cking DANGEROUS:

paste one prompt into Grok Bot before bed and it runs a whole NIGHT SHIFT while you sleep

I still don't understand why people close the laptop and let Grok Bot sit idle for 8 hours

the crew, all Grok Bots:
→ chief (23:00): splits tomorrow's goal into jobs
→ research: reads every source, a link on every line
→ builder: drafts the copy, the emails, the page
→ verifier: a DIFFERENT bot. kills anything with one source
→ inbox: sorts your mail into drafts. sends NOTHING
→ you (7am): one card. max 3 things need your yes

you sleep 8 hours. Grok Bot works 8 hours

and nothing ships without your yes

the prompt:

"you are the chief of my night shift. tomorrow's goal: [your goal]. before 7am:
1. split it into jobs, one bot per job: research, builder, verifier, inbox
2. research puts a source on every line
3. the verifier is a different bot. it rejects anything with fewer than 2 sources and sends it back once
4. inbox sorts my mail and drafts replies. it sends nothing
5. never send, buy, post or delete anything without my yes
6. at 7am give me one card: what's done, what failed, max 3 things that need my yes"

what would you hand the night shift first?

[vartekx @vartekxx]

Head of Claude Code, Boris Cherny:

"85% of our engineers are running dozens or hundreds of agents. The way you do it is graph engineering."

In 40 minutes he explains how a single engineer now does the work of a whole team, how far this has gone inside Anthropic and where it goes next.

These 40 minutes are worth more than any $500 agentic course you'll find online.

Watch it, then read the full guide on graph + context engineering below.

[小雪带你了解汽车 @Maya_Jey]

@emmierath 对，这车轴距长、重心压得低，推起来能感觉出来，坐上去反而稳得不像肌肉车；你觉得宽车身会拖累低速操控吗？

[小雪带你了解汽车 @Maya_Jey]

广州今天试Ducati Diavel，像气场压场的御姐，身形低宽，肌肉感特别明显，油门一开又完全不肯装温柔。佛山、东莞、惠州，完整动态体验放主页。

[小雪带你了解汽车 @Maya_Jey]

@adamthornton09 惠州能约，周末过来直接看真车，想不想顺便试驾一把？

[小雪带你了解汽车 @Maya_Jey]

@jacksonslost 低速挪车是有点沉，推着走不太轻松，但真骑起来市区不累，低扭够用不用老拉高转。你主要跑市区哪边？

[小雪带你了解汽车 @Maya_Jey]

@ErikaBowers99 座高确实低，市区红绿灯双脚是能踏实落地的；就是车宽，腿要往外撑一点。你多高？我可以帮你估一下。

[小雪带你了解汽车 @Maya_Jey]

@bakers124 被压着但一拧就不藏，这个反差就是我说的那个意思，你描述得比我还准。

[小雪带你了解汽车 @Maya_Jey]

@rockhey 对，那条线静态照片一平就全没了，得看车动起来侧面那个光影。你也去试过实车吗？

[小雪带你了解汽车 @Maya_Jey]

@cherrishfly14 哈哈说得对，温柔不温柔看右手，低转确实挺听话的，是我油门给太急了。

[小雪带你了解汽车 @Maya_Jey]

@Melissarivera12 侧面确实最能看出来，油箱那块一路压到座垫的线条，照片拍不出那个厚度。你上次在佛山哪一带见到的？

[小雪带你了解汽车 @Maya_Jey]

@KrisShan1 texture matters with bronzer looks muddy natural shades eventually properly creamy cheaper finish honestly experience routine fresh use until test brands depending

[Shinmaboroshi @Shinmaboroshi]

Anthropic just rebuilt Git for the age of AI agents. 7 engineers did it in 5 months.

“Automate the coordination before you fire the coordinators.”

What’s inside:

→ 0:01 – The bottleneck of scale: 2,500 → 30,000+ merges per day
→ 1:25 – Building Scion, a version control system for machines
→ 3:42 – How 7 people migrated the entire monorepo
→ 5:04 – The productivity J-curve & “don’t pave the cow paths”
→ 7:21 – What this means for labor
→ 8:38 – Automating coordination & post-labor economics
Scion runs on 200x less infrastructure than the old Git mirrors.
Writing code is no longer the bottleneck. Human-built systems are.
Watch it today, then find which processes in your company were built for human speed ↓

[小雪带你了解汽车 @Maya_Jey]

@Arindam_1729 sonic changed the whole room tonight honestly